Geodemographic Classification

The Question

“Who lives here, and how does that affect restaurant viability?”

Data Requirements

Dataset	What We Use	Link
Household Income	Primary clustering variable for socioeconomic segments	View →
Age Distribution	Life-stage segmentation (young professionals vs families vs retirees)	View →
Ethnicity	Cultural dining preference clusters	View →
Household Size	Family vs single-person household segments	View →
School Locations	Family-with-children segment indicator	View →
Employment by District	Industry mix defining neighbourhood character	View →
Crime by Type	Safety perception affecting residential desirability	View →

The Method

Geodemographic systems classify areas by resident characteristics. HK doesn’t have UK’s ACORN or Mosaic, but Census data lets us build a proxy.

Source: de Smith, M., Goodchild, M. & Longley, P. — Geospatial Analysis (6th ed.)

Key Demographic Dimensions

When analysing an area for restaurant viability, these dimensions reveal the most about dining behaviour:

Dimension	What to Look For	Restaurant Implication
Age structure	Proportion of working-age vs elderly vs youth	Young professionals = higher dining-out frequency
Household income	Median and distribution spread	Determines viable price points
One-person households	% single-occupancy	Singles eat out more than families
Ethnicity mix	Cultural diversity of area	Signals demand for non-local cuisine
Employment mix	Office worker density	Weekday lunch traffic predictor
Foreign workers	Expat / domestic worker concentration	Distinct dining days and preferences

Temporal Segmentation

A key insight from geodemographic analysis: the same location can have completely different customer profiles by time of day and day of week.

• Weekday daytime: Office workers, time-constrained, price-sensitive
• Weekday evening: Residents, young professionals, occasion-driven
• Weekend: Families, tourists, discovery-driven, higher spend

Tip

Your target customer choice determines which segment you optimize for. Serving multiple segments requires a dual-mode concept. The LLM Agent Simulation tests whether specific personas would actually visit.

How It Connects

This Model	+ Data Source	= Insight
Geodemographics	Census data	Customer segment sizing
+	Competitor types	Gaps in the market
+	Company concept	Product-market fit

Implementation Notes

Implementation vs Theory

This section documents the actual implementation in functions/api/analyze.ts, including calibration decisions and bug fixes. The theoretical formula above is the target; the implementation below is what currently runs.

Current Implementation (2026-03-25)

Local population: Uses pop2km (STPU census zone sum within 2km) when available. District totalPop can be misleading — Islands district has 182K residents but Discovery Bay has near-zero within 2km. If pop2km > 0, it is used for all local assessments.

Classification order (suburban/low-density checked FIRST, before income):

1. density < 5000 AND (localPop < 100K OR pop2km < 30K) → Suburban (sub-classified below)
2. density < 5000 AND totalPop >= 100K → “Industrial/Commercial Hub”
3. density 5K–15K AND income >= 30K → “Affluent Town Centre”
4. density 5K–15K AND income < 30K → “Town Centre Value”
5. income >= 30K AND diverse >= 0.12 AND density >= 15K → “Cosmopolitan Affluent”
6. income >= 25K AND youngPop >= 0.68 → “Young Professional Hub”
7. income >= 22K AND singleRate <= 0.16 → “Family Neighbourhood”
8. income < 20K AND density > 30K → “Dense Value Seeker”
9. otherwise → “Mixed Urban Core”

Suburban sub-classification (when isSuburban is true):

• family >= 0.43 → “Suburban Family” (Tuen Mun, Yuen Long, Tai Po)
• western >= 0.04 → “Expat Suburban” (Sai Kung, Islands, Southern)
• else → “Suburban Destination”

Industrial/Commercial Hub condition: density < 5000 but totalPop >= 100K — catches low-residential-density industrial zones (Kwun Tong, Kwai Chung) that have massive surrounding workforce not reflected in local density.

Cosmopolitan Affluent requires density >= 15K — ensures it applies only to genuinely urban diverse areas (Central, Wan Chai), not low-density expat suburbs.

Changelog

Date	Change	Why
2026-03-25	Added 6 new segments: Industrial/Commercial Hub, Affluent Town Centre, Town Centre Value, Family Neighbourhood, Dense Value Seeker; “Expat Quarter”/“Affluent Enclave” replaced with “Cosmopolitan Affluent"	"Mixed Urban Core” was too broad — caught Kwun Tong, Jordan, Tsuen Wan, Wong Tai Sin with very different profiles
2026-03-25	pop2km (STPU-based) used as primary local population signal when available	District totalPop misleading for sparse districts (Islands, Sai Kung)
2026-03-25	Cosmopolitan Affluent now requires density >= 15K	Discovery Bay (low density, high income, 10% western) was misclassified as cosmopolitan
2026-03-24	Moved density <5000 check BEFORE income-based checks	Sai Kung (density 900, income 29K) was classified as “Young Professional Hub” instead of suburban
2026-03-24	Added 3-way suburban sub-classification (Family/Expat/Destination)	All low-density areas previously fell into a single undifferentiated suburban bucket